With Cliff Young, President Ipsos
Don’t believe the hype. Popular narrative right now has the Democrats “zero for four” when it comes to elections, with much accompanying angst about the state of affairs in the party. But as always, we encourage armchair pundits to remain grounded in the numbers and the bigger picture, and remember that individual isolated elections are not leading indicators of, well, anything!
A data-driven approach to election prediction takes into account the structural context of an election scenario, and the primary explanatory factors for the 2018 midterm elections are purely structural in nature. Our analysis allows us to look towards 2018 with clearer eyes.
If the 2018 midterm elections were held today, the Republicans would lose control of the House by about seven seats. But they shouldn’t necessarily blame Trump: the reasons are almost entirely out of his hands. There is a ‘rhythm’ to politics and control of the House of Representatives; this is something we can explore using a 5-variable regression model to forecast House seats. This model was tested with data from all midterm congressional elections from 1946 to 2006, a total of 17 data points. The specific model inputs include:
1. Presidential approval ratings, taken from Gallup [LINK] from 1946 to 1990 and from multiple poll averages thereafter
2. The number of Republican seats in the previous congress, a lagged variable
3. The difference in poll data on the “Generic Congressional Ballot” question (Republican share minus Democrat share) [LINK]
4. The party of the sitting president
5. The difference in House seats (Republican minus Democrat) as the dependent variable
Our House model, which has a high predictive value, suggests an estimated 27-seat gain by Democrats, translating into a net 7-seat Democrat advantage.
Our model shows that the two strongest drivers are the Generic Congressional Ballot polling and the party of the president. The fact that the president is a Republican speaks to the “structural” nature of the rhythm of House control. The party newly in control of the White House is in some ways automatically disadvantaged at the next midterm election. Any Republican ― not just President Trump ― would notably increase the Republicans’ chances of losing the House. We would add that the debate over healthcare is also hurting Republicans right now. Healthcare is an issue on which Democrats are typically seen as stronger, and the political chaos around ACA / AHCA will naturally hurt Republicans more than Democrats.
There are some notable caveats that may apply here. First, Democratic advantage may be underestimated due to gerrymandering, which has traditionally helped Republicans. Second, this estimate is based on the present moment, including Trump’s current approval ratings [LINK] and data on today’s Generic Congressional Ballot [LINK].
That said, the highly predictive value of this model paired with the immutability of key variables (number of seats and President’s party) suggest that this prediction is a solid one. It is certainly not outside the realm of possibility that Trump’s ratings will remain as low as they currently are. Indeed, they have not improved significantly from the earliest ratings [LINK] taken of him prior to his election as President. Despite Trump’s lack of support, this model does not factor in the President as an individual, so even a change in the President himself (due to impeachment or any other cause) would not change some of the key model inputs.
Just as we predicted Trump would be the Republican nominee before others, we are now predicting the Democrats will win the House in 2018. Of course, we reserve the right to adjust and update this prediction as new data comes to light. But, at this point, it would likely take a significant change in public opinion to shift these numbers back in Republican favor.